Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

One of the main obstacles in applying engineering approaches to biological processes remains dealing with inter- and intra-individual differences. Therefore, it is highly challenging to accurately monitor their individual state (cfr. personalised medicine). The general objective of this PhD is to develop a framework for individualised model-based monitoring for biological processes, as inspired by control engineering concepts. The presented approach addresses four main topics: i) the biological process itself (i.e. bio-process), ii) the process model, iii) model-based features and iv) individualised change detection based on individual thresholds. In order to explore the general objective, six different case studies (cell, embryo, animal, human) were examined: i) individualised monitoring of activity and body weight in the activity-based anorexia rat model, ii) individualised model-based monitoring of interleukin-6 for early detection of infection in pigs, iii) model-based monitoring of heart rate and blood cytokine time series for early detection of infections in critically ill patients, iv) model-based monitoring of mGluR-dependent synaptic plasticity in hippocampal brain slices of rat, v) individualised monitoring of hippocampal theta oscillations and individualised electrical stimulation in the mesencephalic reticular formation for real-time closed-loop suppression of locomotion in rat and vi) individualised model-based monitoring of chicken embryo status during incubation based on eggshell temperature and micro-environmental air temperature. The results showed that the individual bio-processes involved (individual structure, individual dynamics, bio-signals) can be considered as the biological equivalents of clever-designed control engineering components by defining actuator and homeostatic variables for each of the six case studies (case studies i-vi). Although biological processes are known to contain many nonlinearities, compact individual linear models (general Box-Jenkins models) could be used for the specific individualised monitoring applications of the case studies. By using these models we obtained good approximations of the individual bio-process dynamics (case studies ii, iii, iv and vi), since biological systems often show relatively simple responses (expressing the crucial dominant processes that ascertain healthy internal homeostatic or homeodynamic conditions) when exposed to perturbations as illustrated by the bio-processes of the case studies. In addition, we were able to uncover information about the underlying mechanisms/state by applying data-based mechanistic modelling approaches (i.e. case studies iv and vi). Based on the results, we suggest three different model-based features (model parameter changes, changes in model order and changes in the noise model). In addition, more than 20 other generic metrics from the fields of complex systems science, change detection and control engineering were identified that can be used while analysing individual time series (case studies i-vi). This list of metrics can be used for all individual bio-processes in the design of model-based monitoring application. Based on the specific case studies, three possible approaches were proposed for model-based monitoring of bio-processes based on individual thresholds (e.g. case studies v and vi): 1) individual thresholds based on (sub-)population information, 2) individual thresholds based on universal laws and insights from control engineering, complex systems science and biology and 3) individual thresholds based on individual serial baseline measurements, which can be considered as the most individualised way. To conclude, this thesis has led to some innovative individualised monitoring applications based on each of the six specific case studies. Until now the existence of general frameworks for individualised model-based monitoring of biological processes is limited. Each specific case contributed to the development of such general framework inspired by control engineering concepts. The presented general approach could be used in a broad range of application domains, thus stressing the generic power of the suggested framework for individualized model-based monitoring of (complex) bio-processes.

Choose an application

• Reference Manager
• EndNote
• RefWorks (Direct export to RefWorks)

When acute kidney injury (AKI) develops in a critically ill patient, the patient's mortality rate increases. Therefore, monitoring of the renal function is designated in the intensive care unit (ICU). A RIFLE score depicts the stages of possible kidney injury: Risk, Injury, Failure, Loss and End stage. If a patient reaches the Failure stage, renal replacement therapy (RRT) should be initiated within the next 24 hours to decrease the patient's mortality rate. This work is meant to monitor clearance variables, fluid balance and diuresis at the bedside of patients in the ICU. Since early prediction of AKI could reduce the patients' mortality rate and since AKI is quite common in patients at the ICU, the early prediction of possible AKI is very important. Therefore, a model is created to predict future values of the monitored variables. The prediction error of these variables is up to 30%, with a standard deviation of up to 15%. This means that the predictions can give a general idea of what the value of the variable could be in a few hours, but the predictions are not good enough to really rely on them. An algorithm is created to visualise the RIFLE score for the two approaches (urine output and the combination of glomerular filtration rate and creatinine level). This way it is easier to interpret the patient's kidney status. The monitored data could undergo some unforeseen changes. These changes are traced by the error between the actual data and a model of them. If they differ a lot, a large error is obtained. This probably means that the patient undergoes some unpredicted physiological changes. An alarm is set to warn the caregivers at the ICU only if the error is large enough to exceed a certain threshold. This reduces the amount of alarms in the ICU, which is good for the caregivers who suffer from alarm fatigue. Certain unpredictable physiological changes can be detected for patients at the ICU and the prediction of future values gives an insight in what the possible future course will be of the patient's variables.